Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Use when you need to pick high-quality Unsplash images for product/design assets (avatars, headshots, portraits, large website backgrounds, and abstract wallpapers) and output real Unsplash URLs plus practical instructions for producing the right resolutions and aspect ratios (1:1, 4:5, 3:4, 16:9, 9:16).
.claude/skills/mengto-unsplash-asset-images/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-03 | ✗→✓ | ▲ Improved | -4% | 0% |
| case-05 | ✗→✓ | ▲ Improved | -1% | 0% |
| case-17 | ✗→✓ | ▲ Improved | 137% | 0% |
| case-09 | ✗→✓ | ▲ Improved | 14% | 0% |
| case-15 | ✗→✓ | ▲ Improved | 73% | 0% |
Goal: quickly grab good-looking images from Unsplash and deliver them in the right size + ratio.
For each recommendation, output: 1) Unsplash page URL (canonical) 2) Suggested ratios + sizes for the use case
If the user wants a file, instruct them to use the Download button on Unsplash and then crop/resize in their design tool or image pipeline.
Note: do not include Unsplash source or secondary image links; keep only the main photo page URLs.
Pick images with clean face framing + simple backgrounds.
Suggested deliverables:
Aim for shoulders-up framing, neutral backgrounds, “professional but human”.
Suggested deliverables:
Use these when the vibe is “human story”, not “corporate headshot”.
Suggested deliverables:
Pick wide shots with clean negative space and readable gradients.
Suggested deliverables:
Use when you need “brand-safe”, non-specific visuals.
Suggested deliverables:
If the user asks for “best image”:
crop=faces on production URLs)| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-04 | pass→pass | 13,134 | 5,889 | -55% | 1 | 1 | 0% | 2,303 | 2,842 | +23% | 0 | 0 | — |
case-01 | pass→pass | 15,066 | 6,429 | -57% | 1 | 1 | 0% | 3,079 | 2,949 | -4% | 0 | 0 | — |
case-02 | pass→pass | 14,787 | 8,524 | -42% | 1 | 1 | 0% | 2,840 | 3,506 | +23% | 0 | 0 | — |
case-03 | fail→pass | 15,331 | 5,346 | -65% | 1 | 1 | 0% | 2,782 | 2,675 | -4% | 0 | 0 | — |
case-05 | fail→pass | 14,811 | 4,553 | -69% | 1 | 1 | 0% | 2,678 | 2,664 | -1% | 0 | 0 | — |
case-06 | fail→fail | 10,727 | 7,512 | -30% | 1 | 1 | 0% | 2,032 | 3,241 | +59% | 0 | 0 | — |
case-07 | pass→pass | 6,512 | 3,321 | -49% | 1 | 1 | 0% | 1,152 | 2,226 | +93% | 0 | 0 | — |
case-08 | pass→pass | 8,256 | 4,505 | -45% | 1 | 1 | 0% | 1,372 | 2,377 | +73% | 0 | 0 | — |
case-17 | fail→pass | 5,553 | 4,136 | -26% | 1 | 1 | 0% | 1,045 | 2,481 | +137% | 0 | 0 | — |
case-09 | fail→pass | 13,130 | 7,648 | -42% | 1 | 1 | 0% | 2,923 | 3,318 | +14% | 0 | 0 | — |
case-10 | pass→pass | 5,521 | 3,268 | -41% | 1 | 1 | 0% | 1,017 | 2,213 | +118% | 0 | 0 | — |
case-11 | pass→pass | 8,712 | 4,336 | -50% | 1 | 1 | 0% | 1,617 | 2,468 | +53% | 0 | 0 | — |
case-12 | pass→pass | 9,167 | 4,665 | -49% | 1 | 1 | 0% | 1,720 | 2,569 | +49% | 0 | 0 | — |
case-13 | pass→pass | 5,901 | 2,717 | -54% | 1 | 1 | 0% | 1,170 | 2,178 | +86% | 0 | 0 | — |
case-14 | pass→pass | 8,731 | 4,987 | -43% | 1 | 1 | 0% | 1,539 | 2,494 | +62% | 0 | 0 | — |
case-15 | fail→pass | 8,655 | 3,922 | -55% | 1 | 1 | 0% | 1,440 | 2,490 | +73% | 0 | 0 | — |
case-16 | fail→pass | 5,919 | 2,499 | -58% | 1 | 1 | 0% | 1,077 | 2,228 | +107% | 0 | 0 | — |
case-18 | fail→pass | 4,951 | 1,700 | -66% | 1 | 1 | 0% | 963 | 1,903 | +98% | 0 | 0 | — |
case-19 | pass→pass | 9,362 | 5,644 | -40% | 1 | 1 | 0% | 1,554 | 2,539 | +63% | 0 | 0 | — |
case-20 | pass→pass | 10,421 | 6,944 | -33% | 1 | 1 | 0% | 2,280 | 3,166 | +39% | 0 | 0 | — |
case-21 | pass→pass | 6,086 | 6,398 | +5% | 1 | 1 | 0% | 1,124 | 2,770 | +146% | 0 | 0 | — |
case-22 | pass→pass | 9,686 | 7,510 | -22% | 1 | 1 | 0% | 1,935 | 3,120 | +61% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted. The headline lift of +32 percentage points is the difference between those two pass rates over the 22 comparable cases.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.